Image Dehazing Algorithm Optimization And Implementation Based On Depth Extraction And Region Segmentation

In recent years,the air pollution is getting more and more serious,and the haze weather is also becoming more frequent.They not only seriously affected the visibility,but also makes the images taken by the daily imaging equipment degenerated severely.The outdoor computer vision applications is also hampered hardly..Therefore,the image dehazing algorithm has become a very urgent need.This paper detailedly analyzed the defect in dark channel prior(DCP)dehazing algorithm,and the limitations of the atmospheric scattering model.The main problems to be solved are:(1)Dark channel prior algorithm is easy to produce the halo artifacts effect at the edge of the depth.When using the edge protection operation to solve the halo effect,all edges is protected because we can't distinguish the type of edges,which leads to estimate transmission unreasonable at edges of depth abrupt change and the rest regions.(2)The dark channel prior is invalid in the sky region,which leads to estimate transmission unreasonable in the sky region.(3)Atmospheric scattering model only consider the single scattering effect and neglect the influence of multiple scattering,there are some limitations in the atmospheric scattering model.Focusing on the above problems,this paper presents the following improvement methods:(1)The problem of estimating transmission unreasonable at edges of depth abrupt change and the rest regions: A method via weighted fusion strategy was proposed to estimate transmission,through obtaining the depth information in the foggy image,forming the depth information guided map,and further constructing the weighting coefficient so as to obtain the fine transmission estimation result by weightedly fusing the pixel-wise dark channel and the block-wise dark channel.It protects the discontinuity of depth edges,and reduces the impact of local texture edge noise.(2)The problem of estimating transmission unreasonable in the sky region: a segmentation method based on superpixel and feature point detection is used to segment and extract the sky region in foggy image.After successfully segmenting the sky region,the method based on the dark pixel value scaling of the sky region is used to correct the transmission of the sky region,which effectively improves the estimation of the inaccurate transmission in the sky region and avoids the severe halo and color distortion phenomenon.(3)The problem of the limitations of the atmospheric scattering model: using atmospheric point spread function to build a multi-scattering physics model,in order to avoid the complex calculation of the atmospheric point spread function,generalized Gaussian function is used to replace the atmospheric point spread function.Effectively improve the image quality degradation caused by the multi-scattering effect of suspended particles in the air.Finally,the algorithm proposed in this paper is programmed by using VS2010 and OpenCV2 in Windows environment,and a comparison experiment is designed.The experimental results are compared with four related excellent algorithms.From the accuracy of transmission and the quality of final dehazing result and the objective evaluation index are analyzed,which shows that the improved algorithm proposed in this paper is effective.